Caasi

A lightweight, CLI-first orchestration layer for NVIDIA Isaac Sim, Isaac Lab and the wider robotics ecosystem (ROS 2, Nav2, MoveIt 2, ros2_control, ...).

Headless-first. CLI-first. Process-oriented. Visualization-independent.

Documentation: https://a6y3ap.github.io/caasi/

Note: Caasi is an independent open-source project. It is not affiliated with, endorsed by, or an official product of NVIDIA.

Philosophy

• Use existing software. Don't reinvent it. Caasi discovers, configures, launches, connects, monitors and manages tools that already exist.
• Simulation is not visualization. Run headless by default; attach a viewer only when needed.
• Minimum required stack. The Caasi process itself stays lightweight and never imports the heavy Isaac stack, PyTorch or rclpy — it only delegates to them.
• Don't become another ROS. ROS orchestration is delegated to the real ros2 tooling.
• Unix-friendly. Small commands, --json output, sensible exit codes.

Install

Requires Python 3.10+ on Linux. The Isaac stack, ROS 2 and PyTorch are discovered, not installed, by Caasi.

pip install caasi

Or from source:

git clone https://github.com/a6y3ap/caasi.git
cd caasi
pip install -e .              # add pytest as well: pip install -e ".[dev]"

Either way this installs the caasi command. Shell completion is available:

caasi --install-completion

Quickstart

caasi doctor                    # diagnose the whole environment
caasi doctor --component nvidia # one section only
caasi gpu status                # GPU overview via nvidia-smi
caasi system status             # OS / CPU / RAM / disk summary
caasi info                      # CLI + configured/detected ecosystem versions
caasi config show               # effective configuration
caasi config set tools.isaacsim.default "6.0"

Every command supports --help. Data commands support --json for scripting:

caasi gpu status --json | jq '.gpus[0]["memory.used"]'
if caasi doctor --quiet; then echo "environment ready"; fi

Simulations, training jobs and ROS launches start detached and are tracked as runs — the CLI returns immediately and the work survives your terminal:

caasi sim run experiments/wave.yaml
caasi run list
caasi logs latest --follow      # Ctrl+C stops following, the run keeps going
caasi run status latest

Command surface

More examples:

caasi train experiments/ant.yaml --steps 500000 --envs 4096
caasi ros status && caasi nav launch --map maps/warehouse.yaml
caasi container run nvcr.io/nvidia/isaac-sim:5.1.0 ./runheadless.sh
caasi remote run gpu-box python train.py --steps 1000

Configuration

Global config lives at ~/.config/caasi/config.yaml; a project-local ./caasi.yaml is merged on top. Override the global file with CAASI_CONFIG or --config; CAASI_LANG overrides the language.

language: en
tools:
  isaacsim:
    default: "6.0"
    versions:
      "6.0": { path: /opt/isaac-sim-6.0 }
      "5.1": { path: /opt/isaac-sim-5.1 }
remotes:
  gpu-box:
    host: 10.0.0.5
    user: robot
    port: 2222
    identity: ~/.ssh/id_ed25519
    path: ~/experiments

Manage it with caasi config show|get|set|path|tools.

Examples

examples/navigation.yaml is a commented experiment config — the input format shared by sim run, train, benchmark start and dataset generate:

caasi sim run examples/navigation.yaml --dry-run   # print the launch command, start nothing

Tests

pytest

Documentation

Read it at https://a6y3ap.github.io/caasi/ or open docs/index.html directly in a browser.

License

MIT © a6y3ap — see LICENSE.